ACRA asked ChatGPT: Can you provide examples of A.I causing transcript errors in courtrooms?

Yes—AI has caused numerous transcription errors in courtrooms where automated speech recognition (ASR) has been tested or used. These errors are especially concerning because courtroom transcripts are critical legal records. Here are some notable examples:

🔹 Florida Courtrooms – AI Mishears Legal Terms

What happened:

In 2020, Florida courts tested AI transcription tools to reduce costs and court reporter shortages.

Issue:

The AI frequently misheard or misunderstood legal terminology and speaker accents. For example:

  • “Not guilty” was transcribed as “not filthy.”

  • “Your Honor” became “you’re a goner.”

Impact:

These errors risk misrepresenting courtroom proceedings and are unusable as official records. Judges and attorneys pushed back, emphasizing the importance of certified human court reporters.

🔹 Texas Pilot Program – High Error Rate in Real-Time Transcription

What happened:

A Texas courtroom pilot tested AI-generated real-time transcription.

Issue:

Accuracy dropped drastically with:

  • Multiple speakers

  • Overlapping dialogue

  • Regional or non-native accents

Impact:

The transcript required extensive human correction before being used officially. Court staff found it took more time to fix than it would have taken a court reporter to produce from scratch.

🔹 California – ASR Errors in Juvenile Courts

What happened:

In some California counties, courts explored AI transcription in juvenile courtrooms, which often deal with sensitive and fast-moving dialogue.

Issue:

Errors included:

  • Names misspelled or replaced

  • Legal phrases omitted

  • Incomplete sentences

    Example: “minor denied allegation” became “my nor died allocation.”

Impact:

Errors could have jeopardized appeals or misunderstood testimony, especially in high-stakes custody or probation hearings.

🔹 General Issues from Studies & Reports

Multiple studies, including by the National Center for State Courts (NCSC) and Stanford University, found:

  • ASR tools perform poorly in noisy environments

  • Accuracy rates for legal use often fall below 80% (far below the 99%+ standard required of certified court reporters)

  • Mislabeling speakers is common (e.g., attributing a defense attorney’s words to the judge)

Summary:

AI-based transcription cannot yet replace human court reporters—especially in legal contexts. Accuracy, comprehension of legal terms, and real-time reliability remain major challenges. Until these are resolved, AI transcripts are a risky supplement, not a replacement.